* do not pickle cached hash of arrays

* no clue why this works

* remove unneeded changes

* add an explanation

* add simple test

* add another test

* copy getstate/setstate implementation from dataclasses

* typo?

* DictOfNamedArrays: better repr, keys

* add tags to repr

* make keys() abstract

* add to doc

* fix lint

* remove invalid requirements.txt pytools version specification

---------

Co-authored-by: Andreas Klöckner <inform@tiker.net>

* change attrs to dataclasses

* post-merge fix

* more fixes

* switch to array_dataclass

* use _augment_array_dataclass

* restore attribute validation

* mypy?

* Disable pylint invalid-field-call

---------

Co-authored-by: Andreas Kloeckner <inform@tiker.net>

This reverts commit 3c90e784.

This reverts commit a0da6dc9.

* use numpy 2 type promotion directly

* fix lint

* add comment & test regarding np.result_type with pytato arrays

* adjust README

* revert old_numpy test adjustment (tested manually with numpy 1.26)

* remove comment

* Add new propagation logic

* Fix failing tests

* Actually fix failing tests

* linter fixes + other fixes

* use pytools to generate propagation graph

* add pytools 2024.1.14 as dependency

* add correct pytools version to requirements.txt

* remove dependencies in pyproject.toml, add dependency on pytools 2024.1.14 to setup.py

* IgnoredForPropagationTag -> AxisIgnoredForPropagationTag; update docs

* change wording

---------

Co-authored-by: Andreas Klöckner <inform@tiker.net>

* First shot at implementing 'F' ordered array reshapes

* Remove restriction on reshape order

* Refactor to unify paths

* Slight adjustment

* Update comment wording

* Rough draft of old -> new axis mapping

* Absorb new tag propagation

* Revert changes

* don't linearize everything

* pass tests with new reshaped indices getter

* revert change that belongs to another branch

* fix mypy issues

* fix failing arraycontext test

* fix failing flake8 test

* fixes

* remove unnecessary import

* ruff fixes + small change

* Stopped in the middle

* improved index expression generation

* add test for F ordered reshapes

* remove redundant scalar check

* Improve reshape docs

* Goodbye unnecessary variable

* Way more reshape test coverage

---------

Co-authored-by: Andreas Kloeckner <inform@tiker.net>

* add clone_for_callee to CopyMapperWithExtraArgs

* add TransformMapper/TransformMapperWithExtraArgs

* make CopyMapper/CopyMapperWithExtraArgs inherit from TransformMapper/TransformMapperWithExtraArgs

* expand on purpose of TransformMapper in docstring

* Add node counter tests

* CI fixes

* Add comments

* Remove unnecessary test

* Add duplicate node functionality and tests

* Remove incrementation for DictOfNamedArrays and update tests

* Edit tests to account for not counting DictOfNamedArrays

* Fix CI tests

* Fix comments

* Clarify wording and clean up

* Move `get_node_multiplicities` to its own mapper

* Add autofunction

* Linting

* Add Dict typedef and format

* Format further

* Fix CI errors

* Fix wording

* Implement new DAG generator with guaranteed duplicates

* Apply suggestions from code review

Co-authored-by: Matt Smith <mjsmith1@gmail.com>

* Ruff fixes

* remove prints

* Apply suggestions from code review

Co-authored-by: Andreas Klöckner <inform@tiker.net>

* Add explicit bool for count_duplicates

* Update test/testlib.py

Co-authored-by: Andreas Klöckner <inform@tiker.net>

* Seed random

---------

Co-authored-by: Matt Smith <mjsmith1@gmail.com>
Co-authored-by: Andreas Klöckner <inform@tiker.net>
Co-authored-by: Matt Smith <mjsmith6@illinois.edu>

* Remove the index_access_descr from Einsum object.

* Ruff corrections.

* Make the index_to_access_descriptor a cached property that will automatically build a dictionary for strings to access descriptor. Note that this may not be the exact same string as what the user passed in pt.einsum, but will be equivalent.

* Add a comment on the new cached property.

* Fixing comments made by ruff.

* Remove the cached property. Add in a warning to users requesting a subscript string.

* Remove duplicated code in generating a einsum string. The numpy code generator returned a string which contained spaces after the punctuation. It now no longer does.

* Numpy code generator seems to be dependent on the spaces after the punctuation. So, I am adding those back in.

* Deprecate the old function get_einsum_subscript_str, in favor of get_einsum_specification. Update the code to use the new function.

* Visualization needed to be updated as well.

* Update pytato/utils.py

Co-authored-by: Andreas Klöckner <inform@tiker.net>

* Update pytato/utils.py

Co-authored-by: Andreas Klöckner <inform@tiker.net>

* Updates target.

* Update the documentation to be more clear.

* Fixed a typo.

* Remove excess assert statement.

* Ensure there is a unique error message for arguments of string type since we are removing that functionality.

---------

Co-authored-by: Andreas Klöckner <inform@tiker.net>

* add missing *args, **kwargs in WalkMapper.map_call

* memoize clone_for_callee

* remove default CombineMapper map_call implementation

* don't memoize map_function_definition in cached walk mappers

doesn't make sense, since it doesn't return anything

* support calls in InputGatherer

* make NamedCallResult compatible with attrs cache_hash=True

* enable cache_hash on FunctionDefinition

* enable calls in DirectPredecessorsGetter

* memoize Call creation

* make NamedCallResult.call a property

* remove redundant NamedCallResult.name (already defined in NamedArray)

* memoize NamedCallResult creation

* fix docstring

* remove non-argument placeholder check

now done in arraycontext

* fix equality for FunctionDefinition

* add FIXME

* attempt to fix doc warning

* don't construct NamedCallResult directly

* fix mapper method name in UsersCollector

* add FIXME

* Revert "add FIXME"

This reverts commit 72bf01daf4757c72560c5eaf937377b57ce6b07d.

* Revert "remove non-argument placeholder check"

This reverts commit 73478bc4ac16877f946ce868317d9d1d10fb2ca6.

* add some more missing *args, **kwargs to WalkMapper

* remove some unnecessary *args, **kwargs

* add get_func_def_cache_key to walk mappers to correctly handle function caching when extra arguments are present

* undo memoizing Call creation

* don't use regular dict for function call results

* fix type annotation for function result

* undo memoizing clone_for_callee

doesn't avoid retraversal when the same function is encountered inside the bodies of two different
functions

* add SizeParamGatherer.map_call

default map_call implementation was removed from CombineMapper, so it needs to be here

* Revert "add get_func_def_cache_key to walk mappers to correctly handle function caching when extra arguments are present"

This reverts commit 45eb68d54d6b0e74776fa9bde8d41c7f798b8c49.

* Revert "don't memoize map_function_definition in cached walk mappers"

This reverts commit c0e70cdc8821ae7b14c31acf38e996cdf6bd5896.